Samsung and SK Hynix "adjust strategy": New storage factory production plan advanced

Faced with the surge in demand for memory chips driven by artificial intelligence, South Korea's two major storage giants, Samsung Electronics and SK Hynix, are accelerating the production process of their new wafer fabs, shifting their strategic focus from cautious inventory control to active capacity expansion in order to seize the industry's "super cycle" dividends.

According to the South Korean newspaper Chosun Ilbo, SK Hynix plans to advance the trial operation of its Yongin Phase I wafer fab to February-March next year, starting operations ahead of the completion date. Samsung Electronics will also advance the production start date of its P4 factory in Pyeongtaek from the first quarter of next year to the fourth quarter of this year, moving the production schedule forward by about three months. Both companies will focus on deploying high value-added products in the new production lines, such as high-performance DRAM and HBM (high bandwidth memory).

The backdrop for this round of accelerated capacity expansion is the explosive demand for server chips driven by the global expansion of AI data centers. Data from KB Securities shows that as of February this year, the fulfillment rate of memory chip demand from major customers was only about 60%, with the shortage worsening compared to the fourth quarter of last year. Approximately 70% of Samsung Electronics' memory shipments have already been absorbed by AI data center companies.

The market generally expects the supply tightness to continue until 2027. Citigroup predicts that the supply growth rates for DRAM and NAND flash this year will be 17.5% and 16.5%, respectively, while the demand growth rates will reach 20.1% and 21.4%.

New Production Lines to Start Months Earlier

SK Hynix is constructing the Phase I wafer fab at the Yongin semiconductor cluster, aiming for completion in May next year. The external framework of the project is about half completed, with three of the six clean rooms currently under simultaneous construction. This three-story factory will be six times the size of its Cheongju M15X wafer fab.

According to the Chosun Ilbo, citing informed sources, SK Hynix is preparing to start trial operations ahead of the scheduled completion date, possibly as early as February-March next year. The company plans to quickly install equipment in the clean rooms that are completed first, prioritizing the production of high-performance DRAM (such as DDR5) and HBM products that are in high demand in the AI era.

Samsung Electronics is building the P4 (fourth factory) wafer fab in Pyeongtaek, which was originally scheduled for completion in the first quarter of next year but is now planned to be advanced to the fourth quarter of this year, compressing the production timeline by about three months. Samsung Electronics is flexibly allocating resources between memory and wafer foundry equipment based on market conditions, and P4 is expected to focus on producing high-performance memory that is currently in short supply. It is reported that Samsung Electronics has recently developed a strategy to build a new 10-nanometer sixth-generation (1c) DRAM production line for HBM at the P4 factory, with a monthly production capacity expected to reach 100,000 to 120,000 wafers.

According to a semiconductor industry insider cited by the Chosun Ilbo:

"Korean memory companies are very busy preparing for earlier production times."

Capacity Expansion Still Struggling to Keep Up with Demand Growth

Market research firm Omdia shows that Samsung Electronics' annual DRAM production capacity (measured in wafers) will increase from 7.47 million wafers in 2024 to 8.175 million wafers this year. SK Hynix's capacity during the same period will expand from 5.115 million wafers to 6.39 million wafers. With the early production of new factories, next year's output is expected to grow further The core driving force behind the accelerated expansion of the two companies is the surge in demand for high-performance DRAM for AI data centers. Due to production lines focusing on high value-added HBM chips, the output of general DRAM has relatively decreased, exacerbating supply tightness.

KB Securities pointed out:

"As of February, the intensity of the memory chip supply shortage has worsened compared to the fourth quarter of last year, with the demand satisfaction rate for memory chips from major customers only at 60%. 70% of Samsung Electronics' memory shipments are absorbed by AI data center companies."

Citigroup analyzed that the supply growth rate for DRAM this year is 17.5%, while NAND flash supply growth is 16.5%. In contrast, the demand growth rate for DRAM is expected to reach 20.1%, and the demand growth rate for NAND flash is 21.4%, with demand continuing to exceed supply.

Major market research institutions such as Morningstar and JP Morgan predict that the memory supply shortage will last until 2027. DS Investment Securities stated:

"If the supply growth in 2027 is only 1%, this round of the DRAM cycle will last at least until 2027. The demand for server-centric DRAM is directly related to competitiveness and is difficult to easily cut back, with price increases expected to continue until the third quarter of 2026."

Companies Confirm Significant Increase in Capital Expenditure

Samsung Electronics and SK Hynix both stated at their recent earnings releases that they will increase capital expenditure this year to address the memory shortage. Kim Jae-june, Vice President of Samsung Electronics' Memory Division, stated:

"As AI-related demand is expected to continue, we plan to significantly expand equipment investment by 2026. However, equipment expansion this year and next year will be limited, and the supply shortage may worsen."

This statement highlights the lagging characteristic of semiconductor capacity expansion. Although companies are increasing investment and advancing production timelines, it still takes time from construction to stable mass production, making it difficult to fully alleviate the supply-demand imbalance in the short term.

A semiconductor industry insider explained the strategic intent behind the early trial operation:

"This is to quickly enter the trial operation phase, stabilize the mass production system, and simultaneously send a signal to customers that stable supply can be maintained."